Spaces:
Sleeping
Sleeping
Download src/tabula_engine.py from Akshat20051/Dnn_project: direct link, hf CLI and curl.
- Browser
- Download file 4 kB
-
https://huggingface.co/spaces/Akshat20051/Dnn_project/resolve/main/src/tabula_engine.py
- Command line
-
hf download hf://spaces/Akshat20051/Dnn_project/src/tabula_engine.py
-
curl -L -o tabula_engine.py https://huggingface.co/spaces/Akshat20051/Dnn_project/resolve/main/src/tabula_engine.py
4 kB
| import tabula | |
| import pandas as pd | |
| import numpy as np | |
| def parse_with_tabula(pdf_path): | |
| print(f" [TABULA ENGINE] 🧱 Scanning {pdf_path} using Grid Analysis...") | |
| try: | |
| # 1. THE MAGIC COMMAND | |
| # strict=True tells it to strictly follow grid lines (perfect for bank tables) | |
| # pages='all' scans the whole file | |
| # stream=True forces it to look for whitespace columns (solves the "merged text" bug) | |
| dfs = tabula.read_pdf(pdf_path, pages='all', stream=True, guess=True) | |
| if not dfs: | |
| print(" [ERROR] Tabula found no tables.") | |
| return None | |
| print(f" [TABULA ENGINE] Found {len(dfs)} table fragments. Merging...") | |
| # 2. MERGE ALL PAGES | |
| raw_df = pd.concat(dfs, ignore_index=True) | |
| # 3. CLEAN UP HEADERS (The "Smart" Part) | |
| # Tabula sometimes reads the header as the first row of data. | |
| # We standardise it. | |
| # Create a clean bucket for data | |
| clean_rows = [] | |
| # Iterate through what Tabula found | |
| for idx, row in raw_df.iterrows(): | |
| row_list = row.astype(str).tolist() | |
| row_str = " ".join(row_list).lower() | |
| # Skip junk rows (Headers/Page Numbers) | |
| if "date" in row_str and "balance" in row_str: | |
| continue | |
| if "opening balance" in row_str: | |
| continue | |
| if "page" in row_str and "of" in row_str: | |
| continue | |
| # EXTRACT DATA (Heuristic Mapping) | |
| # We look for the columns based on position (Index 0, 1, etc.) | |
| # This works because Tabula preserves column order perfectly. | |
| try: | |
| # Get the raw values, removing 'nan' | |
| vals = [x for x in row_list if x != 'nan' and x.lower() != 'nan'] | |
| if len(vals) < 3: continue # Skip empty/short rows | |
| # Intelligent Column Guesser | |
| date_val = vals[0] # Date is usually first | |
| desc_val = " ".join(vals[1:-2]) # Description is in the middle | |
| # The last two numbers are usually Amount and Balance | |
| # We need to figure out which is Withdrawal/Deposit | |
| # Clean the money numbers | |
| def clean_money(v): | |
| return float(str(v).replace(',', '').replace('Cr', '').replace('Dr', '')) | |
| money_1 = 0.0 | |
| money_2 = 0.0 | |
| try: money_1 = clean_money(vals[-2]) | |
| except: pass | |
| try: money_2 = clean_money(vals[-1]) # Usually Balance | |
| except: pass | |
| # LOGIC: | |
| # In most statements: [Withdrawal] [Deposit] [Balance] | |
| # If only one money column has a value, Tabula might shift it. | |
| # For this demo, let's assume the detected money is a Withdrawal unless it says "Cr" | |
| withdrawal = money_1 | |
| deposit = 0.0 | |
| # If the description says "Credit" or "Deposit", swap it | |
| if "credit" in row_str or "deposit" in row_str or "cr" in row_str: | |
| deposit = money_1 | |
| withdrawal = 0.0 | |
| clean_rows.append({ | |
| "Date": date_val, | |
| "Description": desc_val, | |
| "Withdrawal": withdrawal, | |
| "Deposit": deposit | |
| }) | |
| except: | |
| continue | |
| df_final = pd.DataFrame(clean_rows) | |
| print(f" [SUCCESS] Tabula extracted {len(df_final)} clean transactions.") | |
| return df_final | |
| except Exception as e: | |
| print(f" [ERROR] Tabula failed: {e}") | |
| return None | |